Fault detection and isolation in robotic manipulator via hybrid neural networks

Abstract

Fault diagnosis systems are important for industrial robots, especially those operated in remote and hazardous environment. Faults in robotic manipulator can cause economic and serious damages. So the Robots need the ability to independently as well as effectively detect and tolerate internal failures in order to continue performing their tasks without the need for immediate human intervention. This saves time and cost involved in repairing the robot. This type of autonomous fault tolerance is also useful for industrial robots in that it decreases down-time by tolerating failures, identifies faulty components or subsystems to speed up the repair process, and prevents the robot from damaging the products being manufactured. So an attempt is made to develop a robust fault detection system to identify and isolate the faults in robot manipulator. In this paper, two artificial neural networks are employed to identify and isolate the faults. A learning architecture, approximation of dynamic behavior of robot manipulator, is used to generate the residual vector, by comparing with actual measured values. First, A multi layer perceptron feed forward network, whose structure is characterized by layered graph, trained with back propagation algorithm is applied to reproduce the dynamic behavior, then counter propagation network which learns a near optimal look uptable approximation to the mapping being approximated. The counter propagation network has the ability to compress a huge amount of data in a few weights and parameters. Simulations employing a SCORBOT ER 5u plus five links robotic manipulator are showed demonstrating that the system can detect and isolate correctly faults that occur in non-trained trajectories. The main contribution of this work is the first application of fault detection and isolation to robot manipulator with non-additive fault.